Scribd
Upload
67 views

Solutions To Exam 1: 1 2 N N A N

ECE 534 exam solutions from Fall 2009 are summarized. For a sequence of random variables with probabilities converging to 1, the indicator random variables converge in probability to 1. For independent random variables with finite fourth moments, convergence in mean square does not necessarily imply convergence of squares in mean square. A random variable that is a power of a uniformly distributed random variable converges almost surely under certain conditions. Convergence in distribution can imply convergence in distribution of translated random variables. Central limit theorem and Chernoff bound are used to approximate binomial probabilities. Minimum mean squared error and conditional expectations are derived for jointly distributed random variables. Properties of jointly Gaussian random variables are used to derive conditional probabilities.

Uploaded by

anthalya
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
67 views

Solutions To Exam 1: 1 2 N N A N

ECE 534 exam solutions from Fall 2009 are summarized. For a sequence of random variables with probabilities converging to 1, the indicator random variables converge in probability to 1. For independent random variables with finite fourth moments, convergence in mean square does not necessarily imply convergence of squares in mean square. A random variable that is a power of a uniformly distributed random variable converges almost surely under certain conditions. Convergence in distribution can imply convergence in distribution of translated random variables. Central limit theorem and Chernoff bound are used to approximate binomial probabilities. Minimum mean squared error and conditional expectations are derived for jointly distributed random variables. Properties of jointly Gaussian random variables are used to derive conditional probabilities.

Uploaded by

anthalya
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 3

ECE 534

Fall 2009

October 13, 2009

Solutions to Exam 1
1. Convergence. In each of the following four parts, you are asked a question about the convergence
of a sequence of random variables. If you say yes, provide a proof and the limiting random variable.
If you say no, disprove or provide a counterexample.
(a) Let A1 , A2 , . . . be a sequence of independent events such that P(An ) 1 as n . Now define a sequence of random variables Xn = 11An , n = 1, 2, . . .. Does Xn converge in probability
as n ?
p.

Ans: We can guess that Xn 1. To prove this, consider P{|Xn 1| }. Clearly P{|Xn 1|
} = 0 n if  > 1, since |Xn 1| cannot exceed 1. Thus it remains to see if this probability converges
to 0 for 0 <  1. For 0 <  1
P{|Xn 1| } = P(Acn ) = 1 P(An ) 0

as n

m.s.

(b) Suppose Xn X as n and E[Xn4 ] < for all n. Does Xn2 necessarily converge in
mean square as n ?
Ans: No. Consider = [0, 1] with the uniform probability measure, and let Xn = n11{[0,1/n4 ]} .
m.s.
2
Then E[X 4 ] = 1 < for all n, and Xn X, with X = 0 a.s., but E[Xn2 Xn1
] = n2 (n 1)2 /n4
2 2
2
1 6= E[X X ] = 0, Thus, by the Cauchy criterion, Xn does not converge in m.s. sense.

(c) Suppose X Unif[1, 1] and Xn = X n . Does Xn converge almost surely as n ?


Ans: Yes. Xn () = X()n 0 for all except that for which X() = 1 or X() = 1, which
a.s.
belong to set of measure 0. Thus Xn 0.
d.

(d) Suppose Xn X, and an is a deterministic sequence such that an a as n . Does


Xn + an necessarily converge in distribution as n ?
Ans: Yes. Using characteristic functions, we have E[ejuXn ] E[ejuX ] for all u R. Thus
E[ej(Xn +an )u ] = ejan u E[ejXn u ] ejau E[ejXu ] = E[ej(X+a)u ]
d.

which means that Xn + an X + a.

2. Let X1 , X2 , . . . be i.i.d. Bernoulli random variables, with


P{Xn = 0} =
Suppose Sn =

3
4

and P{Xn = 1} =

1
4

Pn

i=1 Xi .

(a) Find MX (), the moment generating function of Xn .


Ans: MX () = E[eXn ] = 41 e + 34 .

(b) Use the Central Limit Theorem to find an approximation for P{S100 50} in terms of the
Q() function.
1
3
Ans: = E[Xn ] = 41 and 2 = Var(Xn ) = E[Xn2 ] 2 = 14 16
= 16
. Thus, by the Central Limit
Theorem, (S100 100)/(10) is approximately N (0, 1). Therefore,






S100 100
50 n
50 n
10
P{S100 > 50} = P
>
Q
=Q
10
10
10
3

c
V.
V. Veeravalli, 2009

(c) Now use the Chernoff Bound to show that


P{S100 50}

 50
4
3

Ans: By the Chernoff Bound,



P{S100 50} = P

S100
1

100
2

e100 `(0.5)

where `(0.5) is obtained by maximizing


0.5 ln MX () = 0.5 ln(3 + e ) + ln(4)
Taking the derivative and setting it equal to zero, we obtain that the optimizing satisfies

e
= = ln 3
0.5 =
3 + e
Thus `(0.5) = 0.5 ln 3 ln(3/2) = 0.5 ln 4 0.5 ln 3, and the upper bound follows.

3. (12 pts) Suppose X, Y have joint pdf


(
6x if x, y 0 and x + y 1
fX,Y (x, y) =
0
otherwise
(a) Find E[X|Y ].
Ans: fX,Y (x, y) = 6x 11{0y1} 11{0x1y} . Thus
1y

6xdx 11{0y1} = 3(1 y)2 11{0y1}

fy (y) =
0

and for 0 y 1,
fX|Y (x|y) =

2x
fX,Y (x, y)
=
11{0x1y} .
fy (y)
(1 y)2

Therefore, for 0 y 1,
1y

Z
E[X|Y = y] =

2 (1 y)3
2
= (1 y)
3 (1 y)2
3

xfX|Y (x|y)dx =
0

and E[X|Y ] =

2
(1 Y ).
3

(b) Find the MSE achieved by E[X|Y ], i.e. find the minimum MSE.
Ans: It is easy to see that fx (x) = 6x(1 x)11{0x1} . Thus, the minimum MSE is given by
2

E[X ] E[(E[X|Y ]) ] =
0

4
6x (1 x)dx
9
3

Z
0

3(1 y)4 dy =

3
4
1

=
.
10 15
30

(c) Find E[X|Y


].

Ans: Since E[X|Y ] is linear in Y , E[X|Y


] = E[X|Y ].

c
V.
V. Veeravalli, 2009

4. (14 pts) Suppose X, Y1 , Y2 are zero-mean jointly Gaussian with covariance matrix


4 1 1
X
0
Cov Y1 = 1 1
1 0
1
Y2
(a) Find P{Y1 + Y2 X 10} in terms of the Q() function.
Ans: Let W = Y1 + Y2 X. Then W is Gaussian with E[W ] = 0 and
Var(W ) = E[W 2 ] = E[Y12 ] + E[Y12 ] + E[X 2 ] + 2E[Y1 Y2 ] 2E[XY1 ] 2E[XY2 ] = 1 + 1 + 4 + 0 + 2 + 2 = 10.

Thus P{Y1 + Y2 X 10} = Q( 10 ).

(b) Find E[X|Y1 ] and E[X|Y2 ].


Ans: E[X|Y1 ] = 0 + Cov(X, Y1 )Cov(Y1 )1 (Y 0) = Y1 . Similarly, E[X|Y2 ] = Y2 .

(c) Find fX|Y1 ,Y2 (x|y1 , y2 ).


Ans: We know that given Y1 = y1 , Y2 = y2 , X is Gaussian with mean E[X|Y1 = y1 , Y2 = y2 ], and
variance equal to Cov(e), with e = X E[X|Y1 , Y2 ]. Now, with Y = [Y1 Y2 ]> ,
E[X|Y = y] = 0 + Cov(X, Y )Cov(Y )1 [y] = [1 1]y = y1 y2 .
(Note: we could have concluded this from part (b) using linear innovations.) Similarly,
Cov(e) = Cov(X) Cov(X, Y )Cov(Y )1 Cov(Y , X) = 4 [1 1] [1 1]> = 2
Thus fX|Y1 ,Y2 (x|y1 , y2 ) N (y1 y2 , 2).

(d) Find P({X 2}|{Y1 + Y2 = 0}) in terms of the Q() function.


Ans: The straightforward way to do this problem is to define V = Y1 + Y2 , note that X and V
are jointly Gaussian, find the conditional distribution of X given V using the MMSE approach, and
then compute the above probability. But based on the result of part (c), we can conclude that

fX|V (x|v) N (v, 2). Thus P({X 2}|{Y1 + Y2 = 0}) = P({X 2}|{V = 0}) = Q( 2 ).

(e) Let Z = Y12 + Y22 . Find E[X|Z].


Ans: Note that Cov(X, Z) = E[XY12 ] + E[XY22 ] = 0, since for i = 1, 2,
E[XYi2 ] = E[E[XYi2 |Yi ]] = E[Yi2 E[X|Yi ]] = E[Yi3 ] = 0
Thus

E[X|Z]
= E[X] 0 = 0

c
V.
V. Veeravalli, 2009

You might also like

pFad - Phonifier reborn

Pfad - The Proxy pFad of © 2024 Garber Painting. All rights reserved.

Note: This service is not intended for secure transactions such as banking, social media, email, or purchasing. Use at your own risk. We assume no liability whatsoever for broken pages.


Alternative Proxies:

Alternative Proxy

pFad Proxy

pFad v3 Proxy

pFad v4 Proxy